U.S. patent application number 16/684314 was filed with the patent office on 2021-05-20 for distributed data blocks using storage path cost values.
The applicant listed for this patent is Western Digital Technologies, Inc.. Invention is credited to Arne De Coninck, Sam De Roeck, Stijn Devriendt.
Application Number | 20210149563 16/684314 |
Document ID | / |
Family ID | 1000004488517 |
Filed Date | 2021-05-20 |
United States Patent
Application |
20210149563 |
Kind Code |
A1 |
Devriendt; Stijn ; et
al. |
May 20, 2021 |
Distributed Data Blocks Using Storage Path Cost Values
Abstract
Example distributed storage systems, controller nodes, and
methods provide distributed and redundant data blocks accessed
based on storage path cost values. Storage elements are accessible
through hierarchical storage paths traversing multiple system
components. Data blocks are distributed among the storage elements.
System costs are calculated based on the storage path for reaching
each storage element and a storage path is selected based on a
comparison of the system costs for each storage element. Data
blocks are accessed through the selected storage path.
Inventors: |
Devriendt; Stijn;
(Knesselare, BE) ; De Roeck; Sam; (Ghent, BE)
; De Coninck; Arne; (Erpe-Mere, BE) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
Western Digital Technologies, Inc. |
San Jose |
CA |
US |
|
|
Family ID: |
1000004488517 |
Appl. No.: |
16/684314 |
Filed: |
November 14, 2019 |
Current U.S.
Class: |
1/1 |
Current CPC
Class: |
G06F 3/0661 20130101;
G06F 3/0619 20130101; G06F 3/0653 20130101; G06F 3/0635 20130101;
G06F 3/064 20130101; G06F 3/067 20130101 |
International
Class: |
G06F 3/06 20060101
G06F003/06 |
Claims
1. A system, comprising: a block spreader configured to distribute,
based on a spreading policy, a plurality of data blocks to a
plurality of storage elements distributed among a plurality of
storage nodes, wherein: the plurality of storage elements is
configured to store the plurality of data blocks; and the plurality
of data blocks corresponds to a data object; a storage path
estimator configured to calculate at least one system cost value
for each storage element of the plurality of storage elements
storing the plurality of data blocks; a storage path selector
configured to select, based on a comparison of the at least one
system cost value for each storage element, a subset of the
plurality of data blocks from which the data object is recoverable;
and a block clusterer configured to retrieve the subset of the
plurality of data blocks.
2. The system of claim 1, further comprising: an encoder configured
to disassemble the data object into a predetermined number of
redundant data blocks equal to a number of the plurality of data
blocks; and a decoder configured to assemble the data object from
the subset of the plurality of data blocks.
3. The system of claim 1, wherein: the storage path estimator is
configured to identify, using a hierarchical address, each storage
element of the plurality of storage elements; and the hierarchical
address includes a plurality of hierarchy level identifiers
corresponding to a hierarchical path from at least one controller
node to a selected storage element.
4. The system of claim 3, wherein: the plurality of hierarchy level
identifiers is selected from a storage element identifier, a
storage node identifier, a storage rack identifier, a storage
cluster identifier, a storage site identifier, and a storage system
identifier; and the hierarchical address includes at least three
hierarchy levels.
5. The system of claim 1, further comprising: a system data
collector configured to receive a plurality of system parameters
for each storage element of the plurality of storage elements,
wherein the storage path estimator is further configured to use the
plurality of system parameters to calculate the at least one system
cost value for each storage element.
6. The system of claim 5, wherein: a storage path for each storage
element of the plurality of storage elements is configured to
include a plurality of hierarchy levels corresponding to different
system components; and the system data collector is further
configured to receive the plurality of system parameters for the
storage path for each storage element from different system
components from at least two hierarchy levels of the plurality of
hierarchy levels.
7. The system of claim 5, wherein: the plurality of system
parameters is selected from processing cost, memory cost,
read/write latency, idle time, network delay, and network
bandwidth; and the storage path estimator is further configured to
apply a cost model for a selected set of system parameters from the
plurality of system parameters to calculate the at least one system
cost value for each storage element.
8. The system of claim 5, further comprising: the plurality of
storage nodes including the plurality of storage elements, wherein
the plurality of storage nodes further includes at least one
parameter monitor configured to send at least one system parameter
of the plurality of system parameters to the system data collector;
at least one controller node comprising the block spreader, the
storage path estimator, the storage path selector, the block
clusterer, and the system data collector; and at least one network
device interconnecting the plurality of storage nodes and the at
least one controller node, wherein at least one system parameter of
the plurality of system parameters includes a network cost
associated with communication, between the at least one controller
node and each storage node of the plurality of storage nodes,
through the at least one network device.
9. The system of claim 1, further comprising: a policy engine
configured to receive at least one storage policy selected from a
quality of service policy, a latency target policy, and a power
saving policy, wherein: the storage path estimator is further
configured to apply the at least one storage policy to calculating
the at least one system cost value for each storage element of the
plurality of storage elements; and the storage path selector is
further configured to apply the at least one storage policy to
selecting the subset of the plurality of data blocks from which the
data object is recoverable.
10. A computer-implemented method, comprising: disassembling a data
object into a plurality of data blocks; distributing, based on a
spreading policy, the plurality of data blocks to a plurality of
storage elements distributed among a plurality of storage nodes;
calculating at least one system cost value for each storage element
of the plurality of storage elements storing the plurality of data
blocks; selecting, based on a comparison of the at least one system
cost value for each storage element, a subset of the plurality of
data blocks from which the data object is recoverable; retrieving
the subset of the plurality of data blocks; and assembling the data
object from the subset of the plurality of data blocks.
11. The computer-implemented method of claim 10, further
comprising: identifying, using a hierarchical address, each storage
element of the plurality of storage elements, wherein the
hierarchical address includes a plurality of hierarchy level
identifiers corresponding to a hierarchical path from at least one
controller node to a selected storage element.
12. The computer-implemented method of claim 11, wherein: the
plurality of hierarchy level identifiers is selected from a storage
element identifier, a storage node identifier, a storage rack
identifier, a storage cluster identifier, a storage site
identifier, and a storage system identifier; and the hierarchical
address includes at least three hierarchy levels.
13. The computer-implemented method of claim 10, further
comprising: receiving a plurality of system parameters for each
storage element of the plurality of storage elements, wherein
calculating the at least one system cost value for each storage
element uses the plurality of system parameters.
14. The computer-implemented method of claim 13, wherein: a storage
path for each storage element of the plurality of storage elements
is configured to include a plurality of hierarchy levels
corresponding to different system components; and the plurality of
system parameters for each storage element correspond to: the
storage path for each storage element; and different system
components from at least two hierarchy levels of the plurality of
hierarchy levels.
15. The computer-implemented method of claim 13, wherein: the
plurality of system parameters is selected from processing cost,
memory cost, read/write latency, idle time, network delay, and
network bandwidth; and calculating the at least one system cost
value for each storage element includes applying a cost model for a
selected set of system parameters from the plurality of system
parameters.
16. The computer-implemented method of claim 13, further
comprising: monitoring at least one system parameter of the
plurality of system parameters at each storage node of the
plurality of storage nodes; and monitoring a network cost
associated with communication between at least one controller node
and each storage node of the plurality of storage nodes, wherein
the network cost is at least one system parameter of the plurality
of system parameters used for calculating the at least one system
cost value.
17. The computer-implemented method of claim 10, further
comprising: receiving at least one storage policy selected from a
quality of service policy, a latency target policy, and a power
saving policy, wherein: calculating the at least one system cost
value for each storage element of the plurality of storage elements
includes applying the at least one storage policy; and selecting
the subset of the plurality of data blocks from which the data
object is recoverable includes applying the at least one storage
policy.
18. The computer-implemented method of claim 17, further
comprising: detecting a policy violation based on the comparison of
the at least one system cost value for each storage element not
meeting the at least one storage policy; and reducing, responsive
to detecting the policy violation, incoming storage requests to the
plurality of storage nodes.
19. A system, comprising: a plurality of storage nodes including a
plurality of storage elements; at least one controller node
configured to communicate with the plurality of storage nodes;
means for disassembling a data object into a plurality of data
blocks; means for distributing, based on a spreading policy, the
plurality of data blocks to the plurality of storage elements
distributed among the plurality of storage nodes; means for
calculating at least one system cost value for each storage element
of the plurality of storage elements storing the plurality of data
blocks; means for selecting, based on a comparison of the at least
one system cost value for each storage element, a subset of the
plurality of data blocks from which the data object is recoverable;
means for retrieving the subset of the plurality of data blocks;
and means for assembling the data object from the subset of the
plurality of data blocks.
20. The system of claim 19, further comprising: means for receiving
at least one storage policy selected from a quality of service
policy, a latency target policy, and a power saving policy,
wherein: the means for calculating the at least one system cost
value for each storage element of the plurality of storage elements
includes means for applying the at least one storage policy; and
the means for selecting the subset of the plurality of data blocks
from which the data object is recoverable includes means for
applying the at least one storage policy.
Description
TECHNICAL FIELD
[0001] The present disclosure generally relates to data storage,
and in a more particular example, to distribution of data blocks in
a fault tolerant manner with a predetermined level of redundancy in
a distributed storage system.
BACKGROUND
[0002] Often, distributed storage systems are used to store large
amounts (e.g., terabytes, petabytes, exabytes, etc.) of data, such
as objects or files in a distributed and fault tolerant manner with
a predetermined level of redundancy.
[0003] Some existing object storage systems store data objects
referenced by an object identifier versus file systems. This can
generally allow object storage systems to surpass the maximum
limits for storage capacity of file systems in a flexible way such
that, for example, storage capacity can be added or removed as a
function of the applications, systems, and/or enterprise needs,
while reducing degradation in performance as the system grows. As a
result, object storage systems are often selected for large-scale
storage systems.
[0004] Large-scale storage systems generally distribute the stored
data over multiple storage elements, such as for example solid
state drives, or multiple components such as storage nodes
comprising a plurality of such storage elements. However, as the
number of storage elements in such a distributed storage system
increase, the probability of failure of one or more of these
storage elements increases, as does the probability of failure or
interruption of controller, network, and interface components. To
cope with these issue, distributed storage systems generally use
some level of redundancy at the data, storage node, controller,
interface, client, and network levels.
[0005] As redundancy and complexity in distributed storage systems
increase, it may be an increasing challenge to allocate storage
resources, including processing, memory, and network resources to
specific storage operations. For any given storage request, there
may be a number of storage nodes containing the target data and
each node may be reached by a different data path with different
resource constraints. Further, the dynamic nature of the storage
environment means that the costs associated with each path or
resource may vary based on load, service interruptions, storage
management activities, etc. Differing system configurations,
applications, and organization may also have differing standards on
how cost, time, reliability, and other factors are prioritized when
allocating storage resources.
[0006] As large-scale storage systems scale and create distributed
storage systems supporting storage requests across diverse storage
paths, reliable and efficient implementations for dynamically
selecting storage paths based on their associated costs may be
needed.
SUMMARY
[0007] Various aspects for distributed and redundant data block
storage, particularly, using storage path cost values to determine
where data blocks should be stored and accessed are described.
[0008] One general aspect includes a system that includes: a block
spreader configured to distribute, based on a spreading policy, a
plurality of data blocks to a plurality of storage elements
distributed among a plurality of storage nodes; a storage path
estimator configured to calculate at least one system cost value
for each storage element of the plurality of storage elements
storing the plurality of data blocks; a storage path selector
configured to select, based on a comparison of the at least one
system cost value for each storage element, a subset of the
plurality of data blocks from which a data object is recoverable;
and a block clusterer configured to retrieve the subset of the
plurality of data blocks. The plurality of storage elements is
configured to store the plurality of data blocks and the plurality
of data blocks corresponds to the data object.
[0009] Implementations may include one or more of the following
features. The system may further include: an encoder configured to
disassemble the data object into a predetermined number of
redundant data blocks equal to a number of the plurality of data
blocks; and a decoder configured to assemble the data object from
the subset of the plurality of data blocks. The storage path
estimator may be configured to identify, using a hierarchical
address, each storage element of the plurality of storage elements,
and the hierarchical address may include a plurality of hierarchy
level identifiers corresponding to a hierarchical path from at
least one controller node to a selected storage element. The
plurality of hierarchy level identifiers may be selected from a
storage element identifier, a storage node identifier, a storage
rack identifier, a storage cluster identifier, a storage site
identifier, and a storage system identifier. The hierarchical
address may include at least three hierarchy levels. The system may
further include a system data collector configured to receive a
plurality of system parameters for each storage element of the
plurality of storage elements. The storage path estimator may be
further configured to use the plurality of system parameters to
calculate the at least one system cost value for each storage
element. A storage path for each storage element of the plurality
of storage elements may be configured to include a plurality of
hierarchy levels corresponding to different system components. The
system data collector may be further configured to receive the
plurality of system parameters for the storage path for each
storage element from different system components from at least two
hierarchy levels of the plurality of hierarchy levels. The
plurality of system parameters may be selected from processing
cost, memory cost, read/write latency, idle time, network delay,
and network bandwidth. The storage path estimator may be further
configured to apply a cost model for a selected set of system
parameters from the plurality of system parameters to calculate the
at least one system cost value for each storage element. The system
may further include: the plurality of storage nodes including the
plurality of storage elements, where the plurality of storage nodes
further includes at least one parameter monitor configured to send
at least one system parameter of the plurality of system parameters
to the system data collector; at least one controller node
including the block spreader, the storage path estimator, the
storage path selector, the block clusterer, and the system data
collector; and at least one network device interconnecting the
plurality of storage nodes and the at least one controller node,
where at least one system parameter of the plurality of system
parameters includes a network cost associated with communication,
between the at least one controller node and each storage node of
the plurality of storage nodes, through the at least one network
device. The system may further include a policy engine configured
to receive at least one storage policy selected from a quality of
service policy, a latency target policy, and a power saving policy.
The storage path estimator may be further configured to apply the
at least one storage policy to calculating the at least one system
cost value for each storage element of the plurality of storage
elements. The storage path selector may be further configured to
apply the at least one storage policy to selecting the subset of
the plurality of data blocks from which the data object is
recoverable.
[0010] Another general aspect includes a computer-implemented
method that includes: disassembling a data object into a plurality
of data blocks; distributing, based on a spreading policy, the
plurality of data blocks to a plurality of storage elements
distributed among a plurality of storage nodes; calculating at
least one system cost value for each storage element of the
plurality of storage elements storing the plurality of data blocks;
selecting, based on a comparison of the at least one system cost
value for each storage element, a subset of the plurality of data
blocks from which the data object is recoverable; retrieving the
subset of the plurality of data blocks; and assembling the data
object from the subset of the plurality of data blocks.
[0011] Implementations may include one or more of the following
features. The computer-implemented method may further include:
identifying, using a hierarchical address, each storage element of
the plurality of storage elements, where the hierarchical address
includes a plurality of hierarchy level identifiers corresponding
to a hierarchical path from at least one controller node to a
selected storage element. The plurality of hierarchy level
identifiers may be selected from a storage element identifier, a
storage node identifier, a storage rack identifier, a storage
cluster identifier, a storage site identifier, and a storage system
identifier. The hierarchical address may include at least three
hierarchy levels. The computer-implemented method may further
include: receiving a plurality of system parameters for each
storage element of the plurality of storage elements, where
calculating the at least one system cost value for each storage
element uses the plurality of system parameters. A storage path for
each storage element of the plurality of storage elements may be
configured to include a plurality of hierarchy levels corresponding
to different system components. The plurality of system parameters
for each storage element may correspond to the storage path for
each storage element and different system components from at least
two hierarchy levels of the plurality of hierarchy levels. The
plurality of system parameters may be selected from processing
cost, memory cost, read/write latency, idle time, network delay,
and network bandwidth. Calculating the at least one system cost
value for each storage element may include applying a cost model
for a selected set of system parameters from the plurality of
system parameters. The computer-implemented method may further
include: monitoring at least one system parameter of the plurality
of system parameters at each storage node of the plurality of
storage nodes; and monitoring a network cost associated with
communication between at least one controller node and each storage
node of the plurality of storage nodes, where the network cost is
at least one system parameter of the plurality of system parameters
used for calculating the at least one system cost value. The
computer-implemented method may further include receiving at least
one storage policy selected from a quality of service policy, a
latency target policy, and a power saving policy. Calculating the
at least one system cost value for each storage element of the
plurality of storage elements may include applying the at least one
storage policy and selecting the subset of the plurality of data
blocks from which the data object is recoverable may include
applying the at least one storage policy. The computer-implemented
method may further include: detecting a policy violation based on
the comparison of the at least one system cost value for each
storage element not meeting the at least one storage policy; and
reducing, responsive to detecting the policy violation, incoming
storage requests to the plurality of storage nodes.
[0012] Another general aspect includes a system that includes: a
plurality of storage nodes including a plurality of storage
elements; at least one controller node configured to communicate
with the plurality of storage nodes; means for disassembling a data
object into a plurality of data blocks; means for distributing,
based on a spreading policy, the plurality of data blocks to the
plurality of storage elements distributed among the plurality of
storage nodes; means for calculating at least one system cost value
for each storage element of the plurality of storage elements
storing the plurality of data blocks; means for selecting, based on
a comparison of the at least one system cost value for each storage
element, a subset of the plurality of data blocks from which the
data object is recoverable; means for retrieving the subset of the
plurality of data blocks; and means for assembling the data object
from the subset of the plurality of data blocks.
[0013] Implementations may include one or more of the following
features. The system may further include means for receiving at
least one storage policy selected from a quality of service policy,
a latency target policy, and a power saving policy, where: the
means for calculating the at least one system cost value for each
storage element of the plurality of storage elements includes
applying the at least one storage policy; and the means for
selecting the subset of the plurality of data blocks from which the
data object is recoverable includes applying the at least one
storage policy.
[0014] The various embodiments advantageously apply the teachings
of distributed storage networks and/or systems to improve the
functionality of such computer systems. The various embodiments
include operations to overcome or at least reduce the issues in the
previous storage networks and/or systems discussed above and,
accordingly, are more reliable and/or efficient than other
computing networks. That is, the various embodiments disclosed
herein include hardware and/or software with functionality to
improve the selection of storage paths for storing and accessing
redundantly stored data blocks, such as by using dynamically
generated system cost values for each storage path to enforce
storage policy in accordance with a dynamic cost model.
Accordingly, the embodiments disclosed herein provide various
improvements to storage networks and/or storage systems.
[0015] It should be understood that language used in the present
disclosure has been principally selected for readability and
instructional purposes, and not to limit the scope of the subject
matter disclosed herein.
BRIEF DESCRIPTION OF THE DRAWINGS
[0016] FIG. 1 schematically illustrates an example of a distributed
storage system.
[0017] FIG. 2 schematically illustrates an example client
architecture in which the distributed storage system of FIG. 1 may
operate.
[0018] FIG. 3 schematically illustrates an example of a storage
node of the distributed storage system of FIG. 1.
[0019] FIG. 4 schematically illustrates an example of a controller
node or access node of the distributed storage system of FIG.
1.
[0020] FIG. 5 schematically illustrates some example elements of an
object storage system for the distributed storage system of FIG.
1.
[0021] FIG. 6 schematically illustrates example storage system
components in a hierarchical addressing configuration.
[0022] FIG. 7 illustrates an example method for executing storage
requests based on system cost values.
[0023] FIG. 8 illustrates an example method for dynamically
selecting storage elements in a hierarchical distributed storage
system.
DETAILED DESCRIPTION
[0024] FIG. 1 shows an embodiment of an example distributed storage
system 1. In some embodiments, the distributed storage system 1 may
be implemented as a distributed object storage system which is
coupled to one or more clients 10.1-10.n for accessing data objects
through one or more access nodes 20.1-10.n. The connection between
the distributed storage system 1 and clients 10 could, for example,
be implemented as a suitable data communication network. Clients 10
may host or interface with one or more applications that use data
stored in distributed storage system 1. Such an application could,
for example, be a dedicated software application running on a
client computing device, such as a personal computer, a laptop, a
wireless telephone, a personal digital assistant or any other type
of communication device that is able to interface directly with the
distributed storage system 1. However, according to alternative
embodiments, the applications could, for example, comprise a
suitable file system which enables a general purpose software
application to interface with the distributed storage system 1, an
application programming interface (API) library for the distributed
storage system 1, etc. In some embodiments, access nodes 20 may
include a file interface system for receiving file data requests
from clients 10 according to a file system protocol and access data
in storage nodes 30.1-30.40 using a different storage protocol,
such as an object storage protocol.
[0025] As further shown in FIG. 1, the distributed storage system 1
comprises a plurality of access nodes 20 and a plurality of storage
nodes 30 which may be coupled in a suitable way for transferring
data, for example by means of a conventional data communication
network such as a local area network (LAN), a wide area network
(WAN), a telephone network, such as the public switched telephone
network (PSTN), an intranet, the internet, or any other suitable
communication network or combination of communication networks.
Access nodes 20, storage nodes 30 and the computing devices
comprising clients 10 may connect to the data communication network
by means of suitable wired, wireless, optical, etc. network
connections or any suitable combination of such network
connections. Although the embodiment of FIG. 1 shows only three
access nodes 20 and forty storage nodes 30, according to
alternative embodiments the distributed storage system 1 could
comprise any other suitable number of storage nodes 30 and, for
example, two, three or more access nodes 20 coupled to these
storage nodes 30.
[0026] These access nodes 20 and storage nodes 30 may be built as
general-purpose computers. Alternatively, they may be physically
adapted for arrangement in large data centers, where they are
arranged in modular racks 40.1-40.n comprising standard dimensions.
Exemplary access nodes 20 and storage nodes 30 may be dimensioned
to take up a single unit of such racks 40, which is generally
referred to as 1U. Such an exemplary storage node may use a
low-power processor and may be equipped with ten or twelve high
capacity serial advanced technology attachment (SATA) disk drives
and is connectable to the network over redundant Ethernet network
interfaces. An exemplary access node 20 may comprise
high-performance servers and provide network access to clients 10
over multiple high bandwidth Ethernet network interfaces. Data can
be transferred between clients 10 and such access nodes 20 by means
of a variety of network protocols including hypertext transfer
protocol (HTTP)/representational state transfer (REST) object
interfaces, language-specific interfaces such as Microsoft .Net,
Python or C, etc. Additionally, such access nodes may comprise
additional high bandwidth Ethernet ports to interface with the
storage nodes 30. In some embodiments, HTTP/REST protocols
complying with the Amazon Simple Storage Service (S3) object
storage service may enable data transfer through a REST application
protocol interfaces (API). Such access nodes 20 may operate as a
highly available cluster of controller nodes with one or more
integrated and/or independent interface systems, and provide for
example shared access to the storage nodes 30, metadata caching,
protection of metadata, etc.
[0027] As shown in FIG. 1 several storage nodes 30 can be grouped
together, for example because they are housed in a single rack 40.
For example, storage nodes 30.1-30.4 and 30.7-30.40 each are
respectively grouped into racks 40.1 and 40.n. Access nodes 20 may
be located in the same or different racks as the storage nodes to
which the access nodes connect. A rack may have multiple access
nodes, for example rack 40.1, a single access node as rack 40.n, or
no access nodes (not shown) and rely on an access node in another
rack or storage nodes or clients with built-in access node and/or
controller node capabilities. These racks are not required to be
located at the same location, they are often geographically
dispersed across different data centers, such as for example rack
40.1-40.3 can be located at a data center in Europe, 40.4-40.7 at a
data center in the USA and 40.8-40.10 at a data center in
China.
[0028] FIG. 2 is a block diagram of an example storage network 50
using a client architecture. In some embodiments, distributed
storage system 1 may be embodied in such a storage network 50. As
shown, storage network 50 can include multiple client devices 60
capable of being coupled to and in communication with a storage
network 50 via a wired and/or wireless network 70 (e.g., public
and/or private computer networks in any number and/or configuration
(e.g., the Internet, an intranet, a cloud network, etc.)), among
other examples that may include one client device 60.1 or two or
more client devices 60 (e.g., is not limited to three client
devices 60.1-60.3).
[0029] A client device 60 can be any computing hardware and/or
software (e.g., a thick client, a thin client, or hybrid thereof)
capable of accessing storage system 80 utilizing network 70. Each
client device 60, as part of its respective operation, relies on
sending input/output (I/O) requests to storage system 80 to write
data, read data, and/or modify data. Specifically, each client
device 60 can transmit I/O requests to read, write, store,
communicate, propagate, and/or transport instructions, data,
computer programs, software, code, routines, etc., to storage
system 80. Client devices 60 and storage system 80 may comprise at
least a portion of a client-server model. In general, storage
system 80 can be accessed by client devices 60 and/or communication
with storage system 80 can be initiated by client devices 60
through a network socket (not shown) utilizing one or more
inter-process networking techniques. In some embodiments, client
devices 60 may access one or more applications to use or manage a
distributed storage system, such as distributed storage system 1 in
FIG. 1.
[0030] FIG. 3 shows a schematic representation of an embodiment of
one of the storage nodes 30. Storage node 30.1 may comprise a bus
310, a processor 320, a local memory 330, one or more optional
input units 340, one or more optional output units 350, a
communication interface 360, a storage element interface 370, and
two or more storage elements 300.1-300.10. Bus 310 may include one
or more conductors that permit communication among the components
of storage node 30.1. Processor 320 may include any type of
conventional processor or microprocessor that interprets and
executes instructions. Local memory 330 may include a random access
memory (RAM) or another type of dynamic storage device that stores
information and instructions for execution by processor 320 and/or
a read only memory (ROM) or another type of static storage device
that stores static information and instructions for use by
processor 320. Input unit 340 may include one or more conventional
mechanisms that permit an operator to input information to the
storage node 30.1, such as a keyboard, a mouse, a pen, voice
recognition and/or biometric mechanisms, etc. Output unit 350 may
include one or more conventional mechanisms that output information
to the operator, such as a display, a printer, a speaker, etc.
Communication interface 360 may include any transceiver-like
mechanism that enables storage node 30.1 to communicate with other
devices and/or systems, for example mechanisms for communicating
with other storage nodes 30 or access nodes 20 such as for example
two 1 gigabit (Gb) Ethernet interfaces.
[0031] Storage element interface 370 may comprise a storage
interface such as for example a SATA interface or a small computer
system interface (SCSI) for connecting bus 310 to one or more
storage elements 300, such as one or more local disks, for example
3 terabyte (TB) SATA disk drives, and control the reading and
writing of data to/from these storage elements 300. In one
exemplary embodiment as shown in FIG. 2, such a storage node 30.1
could comprise ten or twelve 3 TB SATA disk drives as storage
elements 300.1-300.10 and in this way storage node 30.1 would
provide a storage capacity of 30 TB or 36 TB to the distributed
storage system 1. According to the exemplary embodiment of FIG. 1
and in the event that storage nodes 30.2-30.40 are identical to
storage node 30.1 and each comprise a storage capacity of 36 TB,
the distributed storage system 1 would then have a total storage
capacity of 1440 TB.
[0032] As is clear from FIGS. 1 and 3 the distributed storage
system 1 comprises a plurality of storage elements 300. As will be
described in further detail below, the storage elements 300, could
also be referred to as redundant storage elements 300 as the data
is stored on these storage elements 300 such that none or a
specific portion of the individual storage elements 300 on its own
is critical for the functioning of the distributed storage system.
Each of the storage nodes 30 may comprise a share of these storage
elements 300.
[0033] As shown in FIG. 3 storage node 30.1 comprises ten storage
elements 300.1-300.10. Other storage nodes 30 could comprise a
similar amount of storage elements 300, but this is, however, not
essential. Storage node 30.2 could, for example, comprise six
storage elements 300.11-300.16, and storage node 30.3 could, for
example, comprise four storage elements 300.17-300.20. As will be
explained in further detail below, the distributed storage system 1
may be operable as a distributed object storage system to store and
retrieve a data object comprising data (e.g. 64 megabytes (MB) of
binary data) and a data object identifier for addressing this data
object, for example, a universally unique identifier such as a
globally unique identifier (GUID). Embodiments of the distributed
storage system 1 may operate as a distributed object storage
system. Storing the data offered for storage by the application in
the form of a data object, also referred to as object storage, may
have specific advantages over other storage schemes such as
block-based storage or file-based storage.
[0034] The storage elements 300 or a portion thereof may be
redundant and operate independently of one another. This means that
if one particular storage element 300 fails its function it can
easily be taken on by another storage element 300 in the
distributed storage system 1. However, as will be explained in more
detail further below, the storage elements 300 are capable of
providing redundancy without having to work in synchronism, as is
for example the case in many well-known redundant array of
independent disks (RAID) configurations, which sometimes even
require disc spindle rotation to be synchronized. Furthermore, the
independent and redundant operation of the storage elements 300 may
allow a suitable mix of types of storage elements 300 to be used in
a particular distributed storage system 1. It is possible to use
for example storage elements 300 with differing storage capacity,
storage elements 300 of differing manufacturers, using different
hardware technology such as for example conventional hard disks and
solid state storage elements, using different storage interfaces
such as for example different revisions of SATA, parallel advanced
technology attachment (PATA), and so on. This may result in
advantages relating to scalability and flexibility of the
distributed storage system 1 as it allows for adding or removing
storage elements 300 without imposing specific requirements to
their design in correlation to other storage elements 300 already
in use in the distributed object storage system.
[0035] FIG. 4 shows a schematic representation of an embodiment of
the access node 20. Access node 20 may include controller node
functions and/or file system interface functions for client systems
using file system protocols to access data stored in data objects
in storage nodes 30. Access node 20 may comprise a bus 210, a
processor 220, a local memory 230, one or more optional input units
240, one or more optional output units 250. In some embodiments,
access node 20 may include object storage management functions,
including object storage interface functions, configuration
management, and/or replication engines.
[0036] Bus 210 may include one or more conductors that permit
communication among the components of access node 20. Processor 220
may include any type of conventional processor or microprocessor
that interprets and executes instructions. Local memory 230 may
include a random access memory (RAM) or another type of dynamic
storage device that stores information and instructions for
execution by processor 220 and/or a read only memory (ROM) or
another type of static storage device that stores static
information and instructions for use by processor 320 and/or any
suitable storage element such as a hard disc or a solid state
storage element. An optional input unit 240 may include one or more
conventional mechanisms that permit an operator to input
information to the access node 20 such as a keyboard, a mouse, a
pen, voice recognition and/or biometric mechanisms, etc. Optional
output unit 250 may include one or more conventional mechanisms
that output information to the operator, such as a display, a
printer, a speaker, etc. Communication interface 260 may include
any transceiver-like mechanism that enables access node 20 to
communicate with other devices and/or systems, for example
mechanisms for communicating with other storage nodes 30 or access
nodes 20 such as for example two 10Gb Ethernet interfaces.
[0037] According to an alternative embodiment, the access node 20
could have an identical design as a storage node 30, or according
to still a further alternative embodiment one of the storage nodes
30 of the distributed object storage system could perform both the
function of an access node 20 and a storage node 30. According to
still further embodiments, the components of the access node 20 as
described in more detail below could be distributed amongst a
plurality of access nodes 20 and/or storage nodes 30 in any
suitable way. According to still a further embodiment, the clients
10 may run an access node 20. According to still further
embodiments, access node 20 may be embodied in separate controller
nodes and interface nodes with or without redundancy among the
controller nodes and/or interface nodes.
[0038] FIG. 5 schematically shows selected modules of an access
node, storage node, and/or controller node with storage management
functions for distributed data block storage and access. Object
storage system 500 may be configured as a node with an architecture
and/or hardware similar to controller nodes and/or storage nodes.
Object storage system 500 may incorporate elements and
configurations similar to those shown in FIGS. 1-4. For example,
object storage system 500 may be configured in a controller node 20
with object storage management functions and metadata store 570
and/or object store 580 may be embodied in one or more storage
nodes 30 in communication with controller node 20.
[0039] Object storage system 500 may include a bus 510
interconnecting at least one communication unit 512, at least one
processor 514, and at least one memory 516. Bus 510 may include one
or more conductors that permit communication among the components
of object storage system 500. Communication unit 512 may include
any transceiver-like mechanism that enables object storage system
500 to communicate with other devices and/or systems. For example,
communication unit 512 may include wired or wireless mechanisms for
communicating with file system clients, other access systems,
and/or one or more object storage systems or components, such as
storage nodes or controller nodes. Processor 514 may include any
type of processor or microprocessor that interprets and executes
instructions. Memory 516 may include a random access memory (RAM)
or another type of dynamic storage device that stores information
and instructions for execution by processor 514 and/or a read only
memory (ROM) or another type of static storage device that stores
static information and instructions for use by processor 514 and/or
any suitable storage element such as a hard disc or a solid state
storage element.
[0040] Object storage system 500 may include or have access to one
or more databases and/or specialized data stores, such metadata
store 570, object store 580, and/or cost model data 590. Databases
may include one or more data structures for storing, retrieving,
indexing, searching, filtering, etc. of structured and/or
unstructured data elements. In some embodiments, metadata store 570
may be structured as reference data entries and/or data fields
indexed by metadata key value entries related to data objects
stored in object store 580. Object store 580 may include data
objects comprised of object data (such as host data), some amount
of metadata (stored as metadata tags), and a GUID. Metadata store
570, object store 580, and/or other databases or data structures
may be maintained and managed in separate computing systems, such
as storage nodes, with separate communication, processor, memory,
and other computing resources and accessed by object storage system
500 through data access protocols. Metadata store 570 and object
store 580 may be shared across multiple object storage systems
500.
[0041] Object storage system 500 may include a plurality of modules
or subsystems that are stored and/or instantiated in memory 516 for
execution by processor 514. For example, memory 516 may include a
storage interface 520 configured to receive, process, and respond
to object data requests from client systems or other nodes in
distributed storage system 1. Memory 516 may include an encoder 530
configured to disassemble data objects into a plurality of data
blocks, such as data shards or symbols for an erasure coding
configuration. Memory 516 may include a block spreader 532
configured to distribute the data blocks among a plurality of
storage nodes or storage elements according to a spreading policy.
Memory 516 may include a block clusterer configured to retrieve a
subset of distributed data blocks for reassembling the data object.
Memory 516 may include a decoder configured to reassemble the data
object from the retrieved data blocks and the encoding
configuration used. Memory 516 may include a storage policy engine
540 configured to define one or more storage policies based on
system cost models for use in selecting storage elements. Memory
516 may include a system data collector 550 configured to receive
system cost parameters for calculation of system costs. Memory 516
may include a storage path estimator 552 configured to calculate
system costs for each storage path. Memory 516 may include a
storage path selector configured to select a storage element based
on comparing system costs for each storage path. In some
embodiments, encoder 530, block spreader 532, block clusterer 534,
decoder 536, storage policy engine 540, system data collector 550,
storage path estimator 552, and/or storage path selector 556 may be
integrated into storage interface 520 and/or managed as separate
libraries or background processes (e.g. daemon) through an API or
other interface.
[0042] Storage interface 520 may include an interface protocol or
set of functions and parameters for storing, reading, and otherwise
managing data requests to an associated distributed storage system
and the corresponding metadata data source for mapping file data to
persistent storage data elements. For example, storage interface
520 may include functions for reading, writing, modifying, or
otherwise manipulating data objects and their respective client or
host data and metadata in accordance with the protocols of an
object storage system.
[0043] In some embodiments, storage interface 520 may include a
plurality of hardware and/or software modules configured to use
processor 514 and memory 516 to handle or manage defined operations
of storage interface 520. For example, storage interface 520 may
include a client request handler 522, a metadata manager 524, a
storage manager 526, and a client response handler 528. For any
given client request, storage interface 520 may receive a client
request through client request handler 522 and determine one or
more operations based on the content of the request. These
operations may include metadata operations handled by metadata
manager 524 and/or object data operations handled by storage
manager 526. The results of these operations may be processed,
formatted, and returned by client response handler 528.
[0044] Client request handler 522 may include an interface and/or
communication event-based condition for receiving object data
requests from one or more clients. For example, client systems may
send an object data request over a network connection and addressed
to object storage system 500 or a port or component thereof. Client
request handler 522 may receive these requests and parse them
according to the appropriate communication and object storage
protocols. For example, client request handler 522 may identify a
transaction identifier, a client identifier, an object identifier
(object name or GUID), a data operation, and additional parameters
for the data operation, if any, from the received message or
messages that make up the object data request.
[0045] Metadata manager 524 may include interfaces, functions,
and/or parameters for creating, modifying, deleting, accessing,
and/or otherwise managing object metadata, such as object metadata
572 stored in metadata store 570. For example, when a new object is
written to object store 580, at least one new metadata entry may be
created in metadata store 570 to represent parameters describing or
related to the newly created object. Metadata manager 524 may
generate and maintain an object data index that enables metadata
manager 524 to locate object metadata within metadata store 570.
For example, metadata store 570 may be organized as a key-value
store and object data index may include key values for data objects
and/or operations related to those objects that are indexed with a
key value that include the object identifier or GUID for each
object. In some embodiments, metadata manager 524 may also manage
object metadata stored in object store 580 with object data 582.
Metadata manager 524 may work in conjunction with storage manager
526 to create, modify, delete, access or otherwise manage object
metadata stored as object tags within object store 580.
[0046] Storage manager 526 may include interfaces, functions,
and/or parameters for reading, writing, and deleting object data
elements in object store 580. For example, object PUT commands may
be configured to write object identifiers, object data 564, and/or
object tags to object store 580. Object GET commands may be
configured to read data from object store 580. Object DELETE
commands may be configured to delete data from object store 580, or
at least mark a data object for deletion until a future garbage
collection or similar operation actually deletes the data or
reallocates the physical storage location to another purpose.
[0047] Other object storage commands may be handled by storage
manager 526, including special functions defined for version
control. Object storage commands processed by storage manager 526
may include parameters for accessing special functions and/or
metadata resources stored with host or client data in the data
objects. Storage manager 526 may work in conjunction with metadata
manager 524 for managing object metadata, including versioning
information. Storage manager 526 may operate in conjunction with
encoder 530, block spreader 532, block clusterer 534, and decoder
536 for encoding and decoding data objects in redundant and
distributed data blocks in accordance with an erasure coding
configuration or other data redundancy method. For example, storage
manager 526 may use encoder 530 and block spreader 532 to write
data to object store 580 and use block clusterer 534 and decoder
536 to read data from object store 580. Encoder 530, block spreader
532, block clusterer 534, and decoder 536 may in turn use storage
policy engine 540, system data collector 550, storage path
estimator 552, and storage path selector 556 to select which
storage elements within a given set of storage elements should be
used for such write and/or read operations.
[0048] Client response handler 528 may include an interface and/or
communication logic for sending response messages, such as result,
status, or error messages, to one or more clients related to object
data requests received. For example, client response handler 528
may wait for processing by metadata manager 524 and/or storage
manager 526 to complete or generate an error, and then provide an
appropriate result or error message to the client system(s) for
each object data request received
[0049] In some embodiments, metadata store 570 may be distributed
across multiple systems, such as a plurality of access systems.
Metadata store 570 and/or portions thereof may be sharded data
stores, wherein the data stores are partitioned into segments
stored in different computing systems. Storage interface 520 may
include the functions for locating and accessing relevant portions
of the sharded data base.
[0050] Encoder 530, block spreader 532, block clusterer 534, and
decoder 536 may comprise an erasure coding engine configured to
store data objects in accordance with a reliability policy which
guarantees a level of redundancy. Encoder 530, block spreader 532,
block clusterer 534, and decoder 536 may individually or
collectively include an interface protocol or set of functions and
parameters for distributing data blocks corresponding to data
objects across designated sets of storage nodes and storage
elements and reassembling those data objects from subsets of those
sets (allow for failures, errors, interruptions, and
optimizations). For example, object storage system 500 may
correctly retrieve a data object distributed among storage elements
even if a number of those storage elements have failed or are
otherwise inaccessible. Encoder 530, block spreader 532, block
clusterer 534, and decoder 536 may implement an erasure coding
configuration to achieve the requirements of a reliability policy
with less overhead than full redundancy. For example, encoder 530,
block spreader 532, block clusterer 534, and decoder 536 may
implement Reed-Solomon codes, fountain codes, or rateless erasure
codes such as online codes, low-density parity-check (LDPC) codes,
raptor codes, and numerous other coding schemes that are
available.
[0051] Encoder 530 may receive a data object from storage manager
526 and disassemble the data object into a predetermined number of
redundant sub blocks, all associated with the same data object
identifier. The predetermined number of data blocks may correspond
to a predetermined multiple of a desired spreading width, based on
the sum of a minimal spreading requirement and a maximum number of
concurrent failures. The maximum concurrent failures tolerance may
correspond to the number of storage elements that store data blocks
for the data object and are allowed to fail concurrently as
determined by the reliability policy. The minimum spreading
requirement may correspond to the minimum number of storage
elements that must be available to reassemble the data object.
Encoder 530 may use an erasure coding configuration to produce the
predetermined number of data blocks. In some embodiments, each data
block may comprise encoded data of equal size, such as the total
data object size divided by a factor equal to a predetermined
multiple of the spreading requirement. Some or all data blocks may
also include decoding data, such as parity data or other decoding
data.
[0052] Block spreader 532 may store the predetermined number of
redundant data blocks on a number of storage elements which
corresponds to the desired spreading width or more. For example,
block spreader 532 may store on each storage element a
predetermined multiple of data blocks such that the data blocks are
evenly distributed among the storage elements. In some embodiments,
these storage elements may be split across module storage nodes or
even multiple storage racks, storage sites, or other layers of a
hierarchical storage architecture. Block spreader 532 may implement
a spreading policy that may include, for example, a number of
storage elements to receive the blocks, the storage pool from which
those storage elements may be selected, and constraints or
algorithms for how they are to be distributed (e.g. evenly across
elements or nodes, randomly within an acceptable range, or using
some other distribution method). In some embodiments, a
hierarchical spreading policy may define additional rules for
spreading data blocks, such as defining a number of sites, racks,
etc. across which the data blocks should be spread. For example,
the spreading policy may include a number of blocks per site,
blocks per rack, blocks per node, blocks per element, etc. In some
embodiments, block spreader 532 may invoke storage path selector
556 to select specific storage elements within a storage pool to
receive one or more data blocks as further described below.
[0053] Block clusterer 534 may receive a request from storage
manager 526 to retrieve a selected data object that has previously
been stored through encoder 530 and block spreader 532. Based on
the data object identifier, block clusterer 534 may locate a
predetermined number of the redundant data blocks that were stored
in the storage elements. Block clusterer 534 does not need to be
able to retrieve all of the encoded data blocks in order to
reassemble the data block, only a subset equal to the minimal
spreading requirement. While the likelihood of outright failures
corresponding to the failure tolerance is extremely low, there may
be service interruptions, delays, resource bottlenecks (processing,
memory, bandwidth, etc.), and other considerations that make some
storage elements easier or less costly to access. Block clusterer
534 may invoke storage path selector 556 to select specific storage
elements among the set of storage elements storing the data blocks
corresponding to the selected data object. For example, storage
path selector 556 may provide a preferred subset for retrieval or a
prioritized list for enabling block clusterer 534 to choose the
least costly storage elements from which to retrieve the subset of
data blocks.
[0054] Decoder 536 may reassemble the selected data object from the
subset of data blocks retrieved by block clusterer 534. For
example, decoder 536 may apply a decoding algorithm and the
decoding data from the data blocks to the encoded data in the data
blocks to reassemble the selected data object from the subset of
data blocks. The resulting data object may be returned to storage
manager 526 and storage interface 520 to respond to a host request
or other storage operation.
[0055] Storage policy engine 540 may include an interface protocol
or set of functions and parameters for receiving, defining, and
otherwise managing storage policies for object storage system 500
and/or a broader distributed storage system. For example, storage
policy engine 540 may include functions for initiating, storing,
modifying, or otherwise manipulating one or more storage policy
parameters in storage policy data 592. The storage policy
parameters may then be used by other components in object storage
system 500 to effect data management operations, such as erasure
coding configurations, spreading policies, and storage path
selection. In some embodiments, storage policy engine 540 may
include or interface with a graphical user interface for enabling
system administrators to view, modify, and otherwise manage storage
policy parameters.
[0056] In some embodiments, storage policy engine 540 may include a
plurality of hardware and/or software modules configured to use
processor 514 and memory 516 to handle or manage defined operations
of storage policy engine 540. For example, storage interface 520
may include a system hierarchy model 542, system parameters 544,
system cost functions 546, and path selection rules 548. For
example, system hierarchy model 542 may define storage paths
available to each storage element, system parameters 544 may define
the available system management data for calculating system costs,
system cost functions 546 may enable numerical modeling of system
costs for each storage path, and path selection rules 548 may
customize selection of relevant hierarchical components, parameter
data, cost functions, comparisons, and priorities for making
storage path selections. Collectively, system hierarchy model 542,
selected system parameters 544, relevant system cost functions 546,
and an active group of path selection rules 548 may define a
dynamic cost model for selecting storage paths and may be reflected
in cost model data 590. This dynamic cost model may be implemented
through system data collector 550, storage path estimator 552, and
storage path selector 556 in support of block spreader 532 and
block clusterer 534.
[0057] System hierarchy model 542 may include data structures,
functions, and interfaces for defining a hierarchical addressing
scheme and related storage paths among components, such as from a
controller node to the many storage elements that make up object
store 580. For example, a storage pool supporting object store 580
may include a large number of storage elements grouped into storage
nodes. The storage nodes may be arranged in racks or clusters using
common power and networking resources, racks and clusters may be
distributed across sites, from different server rooms and
buildings, to geographically dispersed data centers. These layers
of components, from the lowest hierarchical level of the storage
elements themselves, up to one or more top-level controllers,
access, or management nodes (sometimes virtualized) responsible for
allocating storage requests and resulting storage operations, may
define distinct failure points as well as a series or routing
decisions for traversing the storage path from the top level to a
target storage element. In some embodiments, a configuration data
structure reflecting the storage system components, their hierarchy
level, and relationship to other components may be stored in cost
model data 590. In some embodiments, system hierarchy model 542 may
be inherent in the physical and/or logical arrangement of
components and reflected in a hierarchical addressing scheme
regardless of whether the model is otherwise stored in a
configuration map or similar modeling tool. An example system
hierarchical model 542 is further shown and described with regard
to FIG. 6.
[0058] System parameters 544 may include interfaces, functions,
and/or parameters for identifying and selecting system parameters
that may be collected and used for dynamic system cost calculations
and comparisons. For example, system parameters 544 may include
storage parameters, such as capacity, read/write latency, command
queue depths, idle time, buffer space, and operational status.
System parameters 544 may include quantification of computing
resource availability, such as memory resources and processor
resources, including special resources like direct memory access
transfer buffers or hardware encoders, decoders, or parity engines.
System parameters may include network parameters, such as network
delay and network bandwidth, which may be measured and quantified
for communications between each layer of the hierarchy, as well as
aggregated into total values for a complete storage path.
[0059] In any given storage system configuration, a variety of
system parameters may be available. System parameters 544 may
enable the identification and selection from available parameters
for supporting system cost functions 546. For example, system
parameters 544 may be used to configure system data collector 550
to receive data from particular APIs, libraries, daemons, or
services or to query specific data sources for selected parameters.
In some embodiments, system parameters 544 may define one or more
parameter monitors 562 to be instantiated on one or more system
components to collect desired parameter data and provide it to
system data collector 550.
[0060] System cost functions 546 may include interfaces, functions,
and/or parameters for calculating system cost values from one or
more parameters. For example, parameters selected from system
parameters 544 and collected by system data collector 550 may be
variables in one or more cost functions that convert the parameters
into a cost function value associated with a storage path of
interest. System cost functions 546 may include standard cost
models for storage resources, computing resources, and network
resources involved in each storage path. System cost functions 546
may generally reduce or simplify a plurality of component level
parameters (from any level of the hierarchy) into aggregate cost
values for use in comparing costs between storage paths and
executing the logical rules associated with path selection. In some
embodiments, system cost functions 546 may include functions based
on processing cost, memory cost, read/write latency, idle time,
network delay, network bandwidth, and/or other system
parameters.
[0061] Path selection rules 548 may include interfaces, functions,
and/or parameters for selecting and comparing system cost values to
determine a preferred storage path, generally reflecting a storage
policy defined for object storage system 500. For example, path
selection rules 548 may compare the relative merits of different
storage elements and their associated storage paths based on one or
more system constraints that the user cares about. In some
embodiments, path selection rules 548 may correspond to logical
evaluations arranged in ordered decision tree for determining which
storage path should be selected among a plurality of viable storage
paths in a storage pool. For example, a series of logical rules for
path selection may be defined in a rule set data structure to be
executed by a rules engine for comparing possible storage paths and
storage elements.
[0062] In some embodiments, path selection rules 548 may be grouped
into specific policy sets and a user may be able to select a policy
set rather than configuring individual rules. For example, path
selection rules 548 may include a quality of service policy 548.1,
a latency target policy 548.2, or a power saving policy 548.3.
[0063] Quality of service policy 548.1 may include a rule set for
avoiding slow or unavailable storage elements, storage nodes, or
other components during retrieval or storage of data requests. For
example, the storage path cost values may include measured response
time to prior requests sent to each storage element and the
comparison will select the subset with the fastest response times
in order to reduce overall latency for all requests. Quality of
service policy 548.1 may also include an aggressive blacklisting
policy for unresponsive storage elements and other components.
[0064] Latency target policy 548.2 may include a rule set for users
and/or processes that issue storage requests requiring low latency.
For example, the rule set may select the fastest path, similar to
quality of service policy 548.1, but only for users and processes
specifically designated to receive low latency treatment, while
selecting a slower path for users and processes that do not require
low latency (to preserve the faster paths for the low latency
requests). In some embodiments, the low latency targets may be
monitored and if latency for low latency requests exceeds a
threshold value, normal requests may be delayed until the latency
targets are again met for low latency requests.
[0065] Power saving policy 548.3 may include a rule set to optimize
the power efficiency of a distributed storage system (DSS) by
choosing storage paths that allow for storage and retrieval of data
blocks in a power efficient way. For example, when storing a data
object in 18 data blocks, where 8 are redundant, the storage paths
may be selected to place a subset 5 data blocks in storage elements
that can be put off-line to save power, such as a storage node or
rack designated to go into a low power mode once the operation is
complete. The other 13 data blocks remain available with low
latency, of which only 10 are needed for retrieval requests. If
needed, the 5 additional storage elements can be put back online or
otherwise taken out of low power mode. The rule set may include the
logical rules for selecting one or more of the offline storage
paths, such as due to failures among the 13 active data blocks or
exceeding a latency threshold, and thereby initiating those storage
elements to return to active use.
[0066] In some embodiments, one or more policy rule sets in path
selection rules 548 may include rules for storing redundant data
blocks based on processing parameters and the comparative
processing intensity of different data blocks, such as host data
blocks versus parity data blocks. Some data blocks may require more
processing during decoding operations than other data blocks. Some
storage paths may be selected because they are associated with a
low-level processing (such as host data blocks), while other
storage paths may be selected because they are associated
high-level processing (such as parity calculation). For example,
parity data blocks may intentionally be sent to storage elements
that are already associated with a high resource cost (such as high
latency) and would not normally be selected over lower latency
storage paths. Because of the high resource cost, the parity data
block is unlikely to be selected for subsequent storage requests in
favor of the data blocks stored to storage elements with a lower
resource cost (such as low latency). This policy may essentially
reserve the use of high processing cost blocks to situations where
the low processing cost blocks can no longer fulfill the incoming
requests, which may in turn reduce latency and power consumption,
at least until the threshold of needing the high processing cost
blocks is passed.
[0067] System data collector 550 may include interfaces, functions,
and/or parameters for receiving and/or accessing system parameters
for use in selecting storage paths and storage elements for
redundant data blocks. For example, system data collector 550 may
include a service for collecting latency data from storage requests
processed through object storage system 500 and components thereof.
Read and write operations may generate messages and/or log entries
that may be used to calculate elapsed times between the initiating
and completion of read or write requests to storage elements and
these processing times may be aggregated in monitoring data 594 by
system data collector 550.
[0068] In some embodiments, components at various levels of the
hierarchy may be configured to collect performance data regarding
storage operations, workload, processor use, memory use, network
delay, bandwidth, and/or system status. System data collector 550
may be configured to receive system parameter values corresponding
to some or all system parameters through a reporting or messaging
function and/or access to data structures or memory locations
storing these values. System data collector 550 may aggregate
collected information in monitoring data 594 for use by storage
path estimator 552 and/or storage path selector 556. In some
embodiments, some parameter values may be refreshed upon request or
responsive to an action, such as completing the processing of a
prior storage request. Some parameter values may be refreshed
periodically at a time interval that provides acceptable accuracy
for cost model calculation without overburdening the system with
traffic related to parameter value messages or queries.
[0069] In some embodiments, system data collector 550 may be
configured for communication with parameter monitors 562. For
example, parameter monitors 562 may include hardware and/or
software data collectors disposed in individual system components,
such as storage elements, storage nodes, top switches, controllers,
network devices, etc. for collecting parameter data from their
respective components and/or adjacent components. Parameter
monitors 562 may be selectively installed in networked components
and registered with system data collector 550 for aggregating
system parameter values in monitoring data 594. As with some
configurations of system data collector 550, parameter monitors 562
may be configured to monitor specific transactions, mine log data,
and/or access system parameters already generated and stored by
their host system components for aggregation in monitoring data
594.
[0070] Storage path estimator 552 may include interfaces,
functions, and/or parameters for calculating cost values for each
storage element and/or related storage path for use in selecting
storage paths and storage elements for redundant data blocks. For
example, storage path estimator 552 may implement system cost
functions 546 based on system parameters 544 collected by system
data collector 550 and stored in monitoring data 594 to calculate
and store storage path costs 596 for each storage element in a
storage pool (for storage commands) or containing a redundant data
block for a requested data object (for retrieval commands). In some
embodiments, storage path estimator 552 may implement a dynamic
cost model that may aggregate a plurality of system parameters into
a multivariable transfer function for quantifying a system cost
value. For example, rather than directly comparing latency values
measured for each storage element, the relative system cost value
for each storage element may be a weighted composite of a plurality
of cost factors selected from storage costs, computing costs,
and/or network costs. In some embodiments, system hierarchy model
542 may enable the factors to be calculated at each hierarchy level
and selected for each storage path cost based on the hierarchical
addressing scheme. For example, based on a storage element's
hierarchical address, parameter values for the storage element,
storage node, rack, and site may be selected for the storage path
calculation. Note that storage path estimator 552 may use the most
recent system parameters available to estimate the storage path
costs, but the costs may only be estimates and the actual costs
incurred by the resulting storage transaction on the selected
storage paths may be different.
[0071] In some embodiments, storage path estimator may include a
cost calculator 554 that executes system cost functions 546 that
embody a particular cost model. For example, storage path estimator
552 may select a defined set of system parameters and one or more
system cost functions for a path selection task. Storage path
estimator 552 may provide the parameter sets for each storage path
and the selected system cost function(s) to cost calculator 554.
Cost calculator 554 may process each parameter set to generate a
system cost value for each storage element and store the system
cost value and a storage element identifier (such as its
hierarchical address) for each storage path in a table, array, or
similar data structure in storage path costs 596.
[0072] Storage path selector 556 may include interfaces, functions,
and/or parameters for selecting the set of storage elements to
receive redundant data blocks for storage requests and/or the
subset of storage elements from which redundant data blocks are
retrieved for retrieval requests. For example, storage path
selector 556 may use the system cost values calculated by storage
path estimator 552 to evaluate which storage elements should be
used by block spreader 532 or block clusterer 534. Storage path
selector 556 may include cost comparator 558 for comparing the
relative system cost values of the candidate storage elements and
related storage paths. For example, assuming a system cost value
where a lower cost is better than a higher cost, cost comparator
558 may sequentially compare the system cost values from storage
path costs 596 and/or sort those values into an ordered list, so
that the set of storage elements with the lowest cost and meeting
the number of storage elements needed for the operation may be
determined. Storage path selector 556 may then return the selected
set of storage elements to block spreader 532 or block clusterer
534 for use in completing the related operations. Based on a
selected system cost function or dynamic cost model that returns a
relative storage cost value, cost comparator 558 may identify the
lowest cost (or highest cost) storage paths for storage path
selector 556.
[0073] In some embodiments, storage path selector 556 may include a
selection rules engine 560 for making more complex logical
determinations to implement storage policy rules in storage policy
data 592. For example, selection rules engine 560 may execute a set
of path selection rules, such as path selection rules 548, using
storage path costs 596 and a decision tree to select storage
elements. A set of path selection rules may include a plurality of
logical steps that eliminate or select storage elements based on
considerations beyond their relative storage path costs alone. For
example, selection rules may enforce special considerations of
storage elements to use or not use due to load, maintenance status,
or power-saving mode. Selection rules may use a classification of a
data unit as low cost (host data) or high cost (parity data) to
enforce different selection criteria. In some embodiments, storage
path costs 596 may include multiple values for different cost types
(such as storage costs, computing costs, and network costs) for
each storage path and selection rules may determine which values
are used and how they are weighted for cost comparator 558. For
example, a given storage policy may select storage elements with
the lowest storage costs that do not also have computing costs or
network costs over particular thresholds.
[0074] Memory 516 may include additional logic and other resources
(not shown) for processing object data requests, such as modules
for generating, queueing, and otherwise managing object data
requests. Processing of an object data request by storage interface
520 may include any number of intermediate steps that yield at
least one data request to the distributed storage system.
[0075] FIG. 6 shows an example hierarchical model of a distributed
storage system 600, similar to distributed storage system 1 and its
components, as described in FIGS. 1-4. The various levels 604 of
the hierarchical configuration of the components of distributed
storage system 600 form a tree structure and the path from
controller node 602 to each storage element 620 forms a unique
storage path associated with that storage element.
[0076] At the first hierarchy level 604.1, storage elements 620 are
grouped into first components 610.1, such as storage nodes 646. In
the example shown, each storage node component 610.1 may include
ten storage elements 620, such as drives, and the 1200 storage
elements correspond to 120 distinct storage nodes. In some
embodiments, storage elements 620 may be communicatively connected
to storage nodes 646 over their respective storage interfaces
648.
[0077] At the second hierarchy level 604.2, storage elements 620
may be further grouped into second components 610.2, such as
storage racks 644. In the example shown, each storage rack
component 610.2 may include six storage node components 610.1 and
the 120 storage nodes may be grouped into 20 distinct storage
racks. In some embodiments, storage nodes 646 may be connected to
storage racks 644 by a backplane 650 including one or more network
connections from a rack switch or similar component to each storage
node.
[0078] At the third hierarchy level 604.3, storage elements 620 may
be further grouped into third components 610.3, such as sites 642.
In the example shown, each site component 610.3, which may
represent a site network configuration and related equipment, may
include ten storage rack components 610.2 and the 20 storage racks
may be grouped into two distinct sites. In some embodiments,
storage racks 644 may be connected into sites 642 by network
devices 652, such as routers, switches, and network management
devices.
[0079] A system manager 640 may be hosted in controller 602 for
overseeing storage requests to distributed storage system 600. In
some embodiments, controller 602 may host elements similar to those
shown in memory 516 of FIG. 5 for managing storage requests and
encoding them to and decoding them from an object data store
comprised of distributed storage elements 620.
[0080] In some embodiments, distributed storage system 600 may use
hierarchical addresses 630 for identifying individual storage
elements. Each storage element 620 may comprise a corresponding
unique storage element identifier that is a hierarchical address.
For example, the hierarchical address of storage element 620.1 is
[1,1,1,1] and the hierarchical address of storage element 620.1200
is [2,10,6,10].
[0081] In some embodiments, each hierarchical address may be
comprised of several component identifiers that identify the
corresponding hierarchical path configuration for each storage
element 620. These component identifiers may each occupy a position
that is a hierarchy level identifier. For example, each
hierarchical address may be an array of four numerals. The left
most or first numeral in the array may be a hierarchy level
identifier and contain a component identifier for components 610.3
in hierarchy level 3. The next numeral in the array moving right
may be a hierarchy level identifier and contain a component
identifier for components 610.2 in hierarchy level 2. The next
numeral on the array moving right again may be a hierarchy level
identifier and contain a component identifier for components 610.1
in hierarchy level one. The right most numeral is a component
identifier for storage elements 620 and may be considered not to be
a hierarchy level identifier or to identify hierarchy level 0, the
storage elements themselves.
[0082] Hierarchical addresses 630 may identify a storage path by
reference to the specific components at each hierarchy level that
data blocks will pass through between controller 602 and each
storage element 620. The example shown includes three hierarchy
levels resulting in a four-number array of hierarchy level
identifiers to describe the address and storage path. Other
hierarchical storage models may include more or fewer hierarchy
levels and the number of identifiers in the address may generally
equal the number of hierarchy levels plus one for the storage
element identifiers themselves. For example, some configurations
may have only two hierarchy levels and others may have four or
more. Example hierarchy level identifiers may include a storage
element identifier, a storage node identifier, a storage rack
identifier, a storage cluster identifier, a storage site
identifier, and a storage system identifier, though other hierarchy
level identifiers and corresponding component identifiers are
possible.
[0083] As shown in FIG. 7, the object storage system 500 may be
operated according to an example method for selecting storage
elements based on system cost values, i.e. according to the method
700 illustrated by the blocks 702-728 of FIG. 7.
[0084] At block 702, a storage request may be received. For
example, a client request handler may receive an object storage
request that references one or more data objects for storage or
retrieval.
[0085] At block 704, monitoring data may be received including one
or more system parameters. For example, a system data collector may
receive updated system parameters for storage elements in a storage
pool relevant to the incoming data request. In some embodiments,
block 704 may operate concurrently to block 702 with data request
handling proceeding in parallel with monitoring system parameters
relevant to one or more system cost values.
[0086] At block 706, whether the received storage request is a read
operation may be evaluated. For example, the storage request
handler may determine whether the storage operation is a put
request (storage or write operation) or a get request (retrieval or
read operation). If yes, the storage request includes a read
operation and method 700 may proceed to block 720. If no, the
storage request includes a write operation and method 700 may
proceed to block 710.
[0087] At block 710, the data object may be disassembled into data
blocks. For example, an encoder may divide the data object into a
predetermined number of redundant data blocks using an erasure
coding configuration.
[0088] At block 712, system cost values may be calculated for each
storage element in a storage pool that will receive the data
object. For example, a storage path estimator may identify each
storage element in the storage pool and calculate an associated
system cost value for the storage path to reach that storage
element.
[0089] At block 714, system cost values may be compared to
determine the relative costs of each storage element. For example,
a storage path selector may compare the system cost values to sort
the storage elements according to their relative costs and enable
them to be grouped by higher or lower costs.
[0090] At block 716, a subset of storage elements in the storage
pool may be selected to receive the data blocks. For example, the
storage path selector may apply a path selection rule to select a
number of storage elements equaling a desired spreading width and
having the lowest relative costs.
[0091] At block 718, the data blocks may be distributed to the
selected subset of storage elements. For example, a block spreader
may store the data blocks to the selected subset of storage
elements based on a spreading policy.
[0092] At block 720, system cost values may be calculated for each
storage element in the set of storage elements containing data
blocks for the data object. For example, the storage path estimator
may identify each storage element with data blocks associated with
the data object identifier and calculate an associated system cost
value for the storage path to reach that storage element.
[0093] At block 722, system cost values may be compared to
determine the relative costs of each storage element. For example,
the storage path selector may compare the system cost values to
sort the storage elements according to their relative costs and
enable them to be grouped by higher or lower costs.
[0094] At block 724, a subset of data blocks equal to a number of
data blocks needed to recover the data object may be selected based
on their corresponding storage elements. For example, the storage
path selector may apply a path selection rule to select a number of
storage elements equaling a minimal spreading requirement and
having the lowest relative costs.
[0095] At block 726, the data blocks may be retrieved from the
selected subset of storage elements. For example, a block cluster
may read the data blocks from the selected subset of storage
elements and provide them to a decoder for assembly.
[0096] At block 728, the data object may be assembled from the
retrieved data blocks. For example, the decoder may use the erasure
coding configuration to decode the data in the data blocks and
reassemble the original data object for the storage request.
[0097] As shown in FIG. 8, the object storage system 500 may be
operated according to an example method for selecting storage
elements for use in executing storage requests, i.e. according to
the method 800 illustrated by the blocks 802-842 of FIG. 8. In some
embodiments, method 800 may operate in conjunction with the example
method 700 of FIG. 7 to support one or more blocks of method
700.
[0098] At block 802, storage elements may be identified using
hierarchical addresses. For example, each storage element may
include an address that identifies the hierarchy of components with
which the storage element may be grouped (e.g. node, rack, site
network, etc.). Erasure coding components, including storage path
estimators and storage path selectors, may identify storage
elements using the hierarchical addresses and advantageously use
the storage path described by the address in selecting which
storage elements to use for a given request.
[0099] At block 804, storage system and component parameters may be
selected to support one or more cost models for the storage
elements. For example, one or more storage policies and related
cost models may define a set of parameters needed to estimate the
relative system costs for each storage path and a storage policy
engine may enable a user to select storage policies and/or cost
models which the storage policy engine may translate into parameter
selections for use by a system data collector.
[0100] At block 806, system parameters may be received along the
hierarchical storage path. For example, parameters may be received
by the system data collector from the storage element, controller,
and/or one or more components along the storage path, using the
hierarchical address to match the storage element with the
associated component parameters.
[0101] At block 808, one or more cost models may be applied to the
relevant parameters. For example, the storage path estimator may
populate the variables in one or more cost functions for a selected
cost model for each of the storage elements and corresponding
storage paths being evaluated.
[0102] At block 810, a system cost value may be calculated for each
storage path. For example, the storage path estimator may process
the cost functions using the parameters for each of the storage
elements and return a relative system cost value for each storage
element identifier or address.
[0103] At block 812, system cost values may be evaluated against
one or more system cost thresholds. For example, a storage policy
may include logical rules requiring that latency, network lag,
queue depths, or other system cost values not exceed defined
thresholds and the storage path selector may evaluate these
thresholds as part of the path selection rules for each storage
operation.
[0104] At block 814, system cost values for each storage element
may be compared to each other. For example, the storage path
selector may compare the system cost values to sort the storage
elements according to their relative costs and enable them to be
grouped by higher or lower costs.
[0105] At block 816, a set of storage elements for the storage
request may be selected. For example, the storage path selector may
apply path selection rules based on both system cost thresholds and
relative system costs between storage nodes to select a number of
storage elements needed for the operation and meeting the criteria
of the storage policy and related cost model and rules set.
[0106] At block 818, the storage request may be executed based on
the selected set of storage elements. For example, the erasure
encoded data blocks for a data object may be stored to the set of
storage elements or retrieved from the storage elements.
[0107] In some embodiments, at blocks 820, 822, and 824, parameters
may be monitored based on the selected parameters of block 804. At
block 820, storage node parameters may be monitored, such as a
read/write latency value received by the system data collector. At
block 822, network parameters may be monitored, such as a network
delay value for each component in the hierarchy received by the
system data collector. At block 824, computing resource parameters
may be monitored, such as a processor usage value or memory usage
value from one or more components received by the system data
collector.
[0108] In some embodiments, at blocks 830 and 832, a storage policy
may be identified for application to the selection of storage
elements. At block 830, the storage policy may be received or
identified, such as by a storage policy engine. At block 832, the
storage policy may be applied, such as by conversion into a set of
cost models and path selection rules for the storage policy that
may be used by the storage path estimator and/or the storage path
selector.
[0109] In some embodiments, at blocks 840 and 842, a storage policy
violation may be detected and enforced with a remedial action. At
block 840, a storage policy violation may be detected. For example,
the storage path selector may determine that no set of storage
elements are available that meet the storage policy, such as due to
system cost values exceeding one or more thresholds or other rules
not being met. At block 842, a remedial action may be taken in
response to the storage policy violation, such as a reduction if
incoming storage requests. For example, responsive to a storage
policy violation related to high request volume causing
unacceptable levels of latency, the storage interface may reduce
the number of incoming requests by denying some portion of them or
redirecting them to another system until the related system cost
values return to acceptable levels.
[0110] While at least one exemplary embodiment has been presented
in the foregoing detailed description of the technology, it should
be appreciated that a vast number of variations may exist. It
should also be appreciated that an exemplary embodiment or
exemplary embodiments are examples, and are not intended to limit
the scope, applicability, or configuration of the technology in any
way. Rather, the foregoing detailed description will provide those
skilled in the art with a convenient road map for implementing an
exemplary embodiment of the technology, it being understood that
various modifications may be made in a function and/or arrangement
of elements described in an exemplary embodiment without departing
from the scope of the technology, as set forth in the appended
claims and their legal equivalents.
[0111] As will be appreciated by one of ordinary skill in the art,
various aspects of the present technology may be embodied as a
system, method, or computer program product. Accordingly, some
aspects of the present technology may take the form of an entirely
hardware embodiment, an entirely software embodiment (including
firmware, resident software, micro-code, etc.), or a combination of
hardware and software aspects that may all generally be referred to
herein as a circuit, module, system, and/or network. Furthermore,
various aspects of the present technology may take the form of a
computer program product embodied in one or more computer-readable
mediums including computer-readable program code embodied
thereon.
[0112] Any combination of one or more computer-readable mediums may
be utilized. A computer-readable medium may be a computer-readable
signal medium or a physical computer-readable storage medium. A
physical computer readable storage medium may be, for example, but
not limited to, an electronic, magnetic, optical, crystal, polymer,
electromagnetic, infrared, or semiconductor system, apparatus, or
device, etc., or any suitable combination of the foregoing.
Non-limiting examples of a physical computer-readable storage
medium may include, but are not limited to, an electrical
connection including one or more wires, a portable computer
diskette, a hard disk, random access memory (RAM), read-only memory
(ROM), an erasable programmable read-only memory (EPROM), an
electrically erasable programmable read-only memory (EEPROM), a
Flash memory, an optical fiber, a compact disk read-only memory
(CD-ROM), an optical processor, a magnetic processor, etc., or any
suitable combination of the foregoing. In the context of this
document, a computer-readable storage medium may be any tangible
medium that can contain or store a program or data for use by or in
connection with an instruction execution system, apparatus, and/or
device.
[0113] Computer code embodied on a computer-readable medium may be
transmitted using any appropriate medium, including but not limited
to, wireless, wired, optical fiber cable, radio frequency (RF),
etc., or any suitable combination of the foregoing. Computer code
for carrying out operations for aspects of the present technology
may be written in any static language, such as the C programming
language or other similar programming language. The computer code
may execute entirely on a user's computing device, partly on a
user's computing device, as a stand-alone software package, partly
on a user's computing device and partly on a remote computing
device, or entirely on the remote computing device or a server. In
the latter scenario, a remote computing device may be connected to
a user's computing device through any type of network, or
communication system, including, but not limited to, a local area
network (LAN) or a wide area network (WAN), Converged Network, or
the connection may be made to an external computer (e.g., through
the Internet using an Internet Service Provider).
[0114] Various aspects of the present technology may be described
above with reference to flowchart illustrations and/or block
diagrams of methods, apparatus, systems, and computer program
products. It will be understood that each block of a flowchart
illustration and/or a block diagram, and combinations of blocks in
a flowchart illustration and/or block diagram, can be implemented
by computer program instructions. These computer program
instructions may be provided to a processing device (processor) of
a general purpose computer, special purpose computer, or other
programmable data processing apparatus to produce a machine, such
that the instructions, which can execute via the processing device
or other programmable data processing apparatus, create means for
implementing the operations/acts specified in a flowchart and/or
block(s) of a block diagram.
[0115] Some computer program instructions may also be stored in a
computer-readable medium that can direct a computer, other
programmable data processing apparatus, or other device(s) to
operate in a particular manner, such that the instructions stored
in a computer-readable medium to produce an article of manufacture
including instructions that implement the operation/act specified
in a flowchart and/or block(s) of a block diagram. Some computer
program instructions may also be loaded onto a computing device,
other programmable data processing apparatus, or other device(s) to
cause a series of operational steps to be performed on the
computing device, other programmable apparatus or other device(s)
to produce a computer-implemented process such that the
instructions executed by the computer or other programmable
apparatus provide one or more processes for implementing the
operation(s)/act(s) specified in a flowchart and/or block(s) of a
block diagram.
[0116] A flowchart and/or block diagram in the above figures may
illustrate an architecture, functionality, and/or operation of
possible implementations of apparatus, systems, methods, and/or
computer program products according to various aspects of the
present technology. In this regard, a block in a flowchart or block
diagram may represent a module, segment, or portion of code, which
may comprise one or more executable instructions for implementing
one or more specified logical functions. It should also be noted
that, in some alternative aspects, some functions noted in a block
may occur out of an order noted in the figures. For example, two
blocks shown in succession may, in fact, be executed substantially
concurrently, or blocks may at times be executed in a reverse
order, depending upon the operations involved. It will also be
noted that a block of a block diagram and/or flowchart illustration
or a combination of blocks in a block diagram and/or flowchart
illustration, can be implemented by special purpose hardware-based
systems that may perform one or more specified operations or acts,
or combinations of special purpose hardware and computer
instructions.
[0117] While one or more aspects of the present technology have
been illustrated and discussed in detail, one of ordinary skill in
the art will appreciate that modifications and/or adaptations to
the various aspects may be made without departing from the scope of
the present technology, as set forth in the following claims.
* * * * *